Mark Bedau (Reed College)
Abstract: The evolution of technology innovation networks is interesting and challenging in part because it is so much more complex than biological evolution. One sign of this complexity is highly variable and very large number of “parents” that an individual innovation have; another sign is that the “traits” possessed by of an individual innovation have a content with rich semantic significance for humans technology designers and users. Patent citation networks are a familiar proxy for studying the evolution of technology, and contemporary text-mining tools provide new opportunities for visualizing and quantifying the hyper-parental evolution of the “content” of technological innovations. We examine the evolution of three complementary kinds of traits of patented innovations: (i) the technology categories, sub-categories, and codes assigned to each invention by human experts (patent examiners), (ii) keywords (tf-idf stems) automatically extracted from the text in each individual patent record, and (iii) nearby clusters in a semantic space automatically constructed from the entire patent record (using word2vec). We see that reconstructing the genealogy of each patented invention and coloring each node with an invention’s traits provides a vivid and powerful visualization of the flow of technology traits through the population of patented inventions. G. Price’s covariance equation for the evolution of traits partitions the evolution of a trait into components corresponding to selection and mutation processes (as well as a third hyper-parental process). Measurements of the Price equation through the patent record reveal the changes in how different evolutionary processes drive various technology sectors (identified by characteristic traits).